Phone and email from one memory

Short version: Customers switch the channel, not the matter. An assistant that knows yesterday's call while reading today's mail saves the second explanation. It may only use that knowledge when the person consented and when a human reads the reply, because a sender address proves nothing.

The typical case

Monday, 4:40 pm: Ms Muster calls the bike shop. The phone assistant notes that her bike should be ready by Friday and saves a callback request with her number. Tuesday, 8:10 am: she writes a mail: "As discussed, please replace the bell too." Without a shared memory the email assistant asks what this is about. With a shared memory the draft knows: call yesterday, bike by Friday, callback open. The reply fits the case, not just the mail.

How the matching works, and how it does not

The assistant connects channels through what a business stores anyway: its leads. A lead has a name, an email address, a phone number. When an address writes that belongs to a lead with a number, the assistant looks up calls and WhatsApp messages from that number in the last 30 days. When a number calls that belongs to a lead with an email address, the phone assistant knows whether a mail from that address is still waiting for a reply.

That is matching, not identity verification. Anyone can send a mail with someone else's sender address. Hence the rule: what the assistant knows from another channel is context for the reply, never proof of who it is talking to.

Three rules that follow

• No content without consent. The content of a call exists only when the caller agreed to the transcript. Otherwise the assistant knows just this: there was a call.
• Automatic replies mention the contact at most. A mail that goes out unread may say "you called us on Monday". It must not repeat what was said on the phone. That stays reserved for drafts a person reads.
• Nothing to the wrong person. If the person writing seems to be someone other than the person who called, the assistant says nothing about the other channel.

The call becomes a draft

The second benefit runs the other way. When a call ends whose number belongs to a known email address and whose caller agreed to the transcript, a draft appears in the mailbox shortly after: "Summary of our phone call", two to four sentences from the customer's point of view, in the caller's language. The assistant never sends it itself. You read, edit or delete. For many businesses this is the follow-up mail that otherwise never gets written.

And if the assistant did not know?

A question you had to answer yourself should not land with you a second time. So a conversation you replied to becomes a knowledge entry with one click: the assistant phrases question and answer in general terms, removes addresses, numbers and references, you check the text and save it. From then on chat, phone and mailbox answer the question alone, because they read the same knowledge base. Today's mail answers next week's chat.

Where all this runs

Anyone handling customer data across several channels wants to know where the AI computes. The workspace shows per mailbox whether the reply model and the search index run in the EU; the mailbox always uses the model you chose for your chatbot. With WebChatAgent you find this under Integrations, Email; the phone assistant uses the same knowledge base and the same leads.